4.2 LGAD for timing
4.2.1 CNM LGAD production run 9088
The first CNM thin LGAD production (run 9088) has been delivered in May 2016.
The production includes a total of 14 wafers, four of them produced on 300µm thick float zone wafers for comparison with previous productions, the others have been made on 100 mm silicon-on-insulator (SOI) with a nominal thickness of 50µm on a 300µm thick support wafer. The effective thickness is reduced to 45µm due to the implantation of the n+ and p+ electrodes in the front and back side respectively.
Each wafer contains different structures, mostly single pad square diodes of active area 1.3×1.3 mm2 and 3.3×3.3 mm2 named respectively LGA and LGB, while the largest part of the wafer is used for arrays of diodes designed for the HGTD and
CT-PPS/TOTEM experiments, see figure 4.13.
Figure 4.13: Picture of a wafer from CNM LGAD production run 9088.
The doping profile of the diodes is similar to the one described in section 4.1. The central region of the diode hosts the n+ electrode and the p+ multiplication layer and is 1 mm wide. It is surrounded by the JTE, that improves the diode voltage capability and defines the electric field at the sensor periphery, a p-stop and a guard ring. The total active area of the diode is defined by the p-stop ring and is 1.3 mm wide, but only the charge collected in the central region will be amplified through the CM avalanche mechanism. The support wafer and the buried oxide layer on the back side are etched to allow to electrically contact thep-type electrode. The backside of the wafer is fully metallized while on the top side of LGA and LGB diodes metal contacts are placed on the guard rings and the central region with an opening to allow laser testing, see figure 4.14.
On the ten SOI wafers the implantation of the multiplication layer has been done with three different doses in order to study devices with different performances in terms of gain and voltage capability. The implantation doses of the multiplica-tion layer were 1.8 (low) for wafers 3 and 4, 1.9 (medium) for wafers 5 to 10 and 2.0·1013cm−2 (high) for wafers 11 and 12. Each wafer also contains diodes without multiplication layer to be used as reference.
The measured samples are described in table 4.2. The samples have been char-acterized with IV and CV measurements on a probe station to verify the voltage capability and the voltage of full depletion. Some of the sensors have been irradiated with thermal neutrons at the TRIGA nuclear reactor of the Joˇzef Stefan Institute (JSI) in Ljubljana to 3 and 10·1014neq/cm2.
4.2. LGAD FOR TIMING 69
(a)
1mm mult.
layer 1.3mm
JTE Guard Ring
(b)
Figure 4.14: Schematic cross section 4.14a and top view 4.14b of the LGA diode in CNM run 9088.
Measured Implantation Fluence
Short name
devices dose
1014neq/cm2
W3-LGA-61 low 0 low,unirr,L1
W3-LGA-71 low 0 low,unirr,L2
W3-LGA-33 low 0 low,unirr,L3
W5-LGA-45 med 0 med,unirr,L1
W5-LGA-81 med 0 med,unirr,L2
W5-LGA-51 med 3 med,3e14,L1
W7-LGA-45 med 3 med,3e14,L2
W5-LGA-43 med 10 med,1e15,L1
W7-LGA-35 med 10 med,1e15,L2
Table 4.2: List of the measured devices with the n+implantation dose and irradiation fluence.
The plots in figure 4.15 show the capacitance (CV) and current (IV) values against the bias voltage. The CV curve is represented as 1/C2to better show where depletion occurs. The slope depends on the doping concentration while the asymptotic value is defined by the sensor thickness. Below∼32 V the multiplication region is slowly being depleted and 1/C2stays at low values. Subsequently, in about 3 V the high resistivity bulk is depleted and the capacitance drops to a value of 3.9 pF, compatible with a diode of 45µm active thickness. The capacitance drop happens with of about 1 V of difference in the bias voltage between the low and medium dose samples, compatible with the different concentration of dopants. From the IVs shown in figure 4.15b it is interesting to note that after irradiation the devices can be operated at significantly larger voltages than before irradiation. This will eventually allow to partially recover the LGAD performance after irradiation.
90Sr Characterization
The response of these devices to beta particles from a90Sr source has been measured at the JSI laboratories in Ljubljana. From these measurements the total collected charge
bias [V]
Figure 4.15: Characteristic CV (a) and IV (b) of the LGA diode in CNM run 9088.
The CV is shown only for one non irradiated sample of each implantation dose, the IV is shown for all the measured samples, at the temperature of 20◦C for the non irradiated and at−10◦C for the irradiated ones.
and therefore the gain evolution with bias voltage of irradiated and non irradiated devices can be extracted.
The sensors were mounted on a dedicated aluminium box with small holes on the top and on the bottom of the sensor and placed below the radioactive source on a cold plate that also has an opening at the position of the sample. A small scintillator, roughly of the size of the DUTs, is placed underneath the cold plate and is used to trigger the readout. Only the electrons crossing the central part of the detector reach the scintillator thanks to the opening in the box.
The cold plate was used to maintain a stable working condition at −10◦C for the irradiated samples and at 20◦C for the non irradiated ones. The signals from the DUT were amplified in two stages by a charge sensitive preamplifier and by a shaping amplifier with a peaking time of 25 ns. The signals were then recorded by an oscilloscope and stored for a later analysis.
In order to have an absolute measure the oscilloscope scale was calibrated with the 59.5 keV photon peak of an americium (241Am) source in a 300µm thick non-irradiated float zone detector.
For each voltage, the amplitude spectrum of the collected waveforms was fitted with the convolution of a Landau and Gaussian distribution to obtain the MPV of the signal amplitude, that was successively converted to the collected charge MPV through a calibration factor. The gain is obtained comparing the charge MPV with the one of an equivalent 45µm thick sensor without charge multiplication layer from the same run, measured to be 2.88 ke−.
The results are summarized in figure 4.16 where we can firstly see that devices with the same implantation dose and fluence behave similarly, confirming the good
4.2. LGAD FOR TIMING 71 reproducibility of the LGAD process. The effect of the different implantation can be compared in the non irradiated devices where at the same bias voltage the medium dose devices have a larger gain than the low dose sample. However, the low dose sample can be operated at a larger bias voltage reaching a maximum gain value similar to the one of the medium dose sample.
Irradiation has the effect of effectively removing the acceptor dopants in the silicon substrate, including the multiplication layer [51, 61]. As a consequence the gain of irradiated devices is suppressed. At a bias value of 200 V the gain is only 4 for the 3·1014neq/cm2 devices and 1 for the 1015neq/cm2 ones, against the gain of 20 that devices with the same multiplication layer have before irradiation. Nevertheless these devices can be operated at higher bias voltage reaching a gain of about 18 at 425 V and 10 at 650 V respectively.
bias [V]
Figure 4.16: Charge collection MPV and gain measured with a 90Sr source.
Beam Test Characterization
The same devices have been tested in the H6B beam line of the CERN-SPS North Area beam test facility with 120 GeV pions for a first measurement of the timing resolution of 45µm LGAD devices.
The sensors were mounted on a Printed Circuit Board (PCB) originally designed for TCT measurements. The bias is applied to the backside through a low-pass RC filter, the central pad is connected via a bond wire to an SMA connector while the guard ring is floating. A broad band amplifier is connected to the SMA and the amplified signal is then recorded by an Agilent Infiniium DSA91204A oscilloscope [62]
with 40 GS/s sampling rate and a band width of 12 GHz. In addition two Silicon Photo-Multipliers (SiPMs) coupled to Cherenkov-light emitters quartz bars have been used as fast (10 ps) timing reference [63]. The quartz bar, with a cross section of
3×3 mm2 and a length of 30 mm, were aligned with the beam direction.
RCE setup outside the fence (3.5 m) for easy resetting without area access
SiPM1 SiPM2
LGADs Amplifiers
Beam
Sensor
Figure 4.17: The LGAD measurement set-up during the beam test in the H6B beam line of CERN-SPS.
The PCB with the LGAD devices were mounted on an aluminium support frame and together with SiPMs fixed to an aluminium base plate, see figure 4.17. The SiPMs were mounted on a mechanically adjustable pedestal to permit the alignment with the LGAD sensors while the base plate was mounted on a remotely controlled movable stage to align the sensors with the pion beam. It was possible to cover the base plate with a styrofoam box to ensure light tightness and thermal insulation during cooling. The cooling was needed for the operation of irradiated devices and was performed with dry ice bricks placed inside the styrofoam box.
Dry ice is solid state carbon dioxide that sublimate at the temperature of−75◦C, the cold gaseous CO2 originated by the sublimation cool down the environment in the styrofoam box. Nevertheless due to a non perfect thermal insulation and the presence of power consuming electronics acting as heat sources, the temperature of the sensors results to be much higher. The irradiated sensors have been measured in two batches, first the ones irradiated to 3·1014neq/cm2 and then the ones irradiated to 1015neq/cm2. The temperature on the sensor was estimated comparing the sensor leakage current to the one measured in a laboratory climate chamber at different temperatures, resulting to be respectively−6◦C and−15◦C.
Different amplifiers were available: the CIVIDEC C2 TCT amplifiers [64], the Particulars TCT amplifiers [65] and the custom-made AFP pre-amplifiers [63]. Inital tests have been carried out to find the amplifier with the best performance in terms of time resolution. The CIVIDEC C2 TCT showed to perform better than the others and have been therefore used to characterize the LGAD devices. Also the oscilloscope settings have been investigated to find an optimal configuration. The bandwidth of
4.2. LGAD FOR TIMING 73 the oscilloscope was varied from 0.5 GHz to 12 GHz, a too low setting would alter the rising edge of the signal while a too high value would introduce high frequency noise without an improvement of the signal rise time. The best compromise was found to be at 1 GHz, set as default value. The vertical scale setting of the oscilloscope was found to influence the noise level and therefore the time resolution, it was set to 50 mV/div and kept constant for all the measurements.
The ambient noise of the beam area and the non optimal sensor assembly on the PCB, with long bond wires and no shielding, led to run-to-run variation of the noise and the signal. The noise level was varying between 3 and 4 mV, while the signal amplitude and integrated charge had variations up to 30−40%. However, the impact on the time resolution was only about 10%.
For each run about ten thousand events were recorded. Two LGAD devices were connected to the oscilloscope together with the two SiPMs, although on some runs only one SiPM was available. The readout was triggered with a fixed threshold on the signal of one of the LGAD devices. Thanks to the large signal over noise ratio (S/N) this was possible without introducing a bias in the signal amplitude and improving the purity of the trigger, since the cross section of the quartz bar was nine times larger than the LGAD active area.
The trigger threshold was typically set to 15÷20 mV. For the runs at the highest voltages this value had to be increased to avoid fake triggers due to micro-discharges in the sensor but since the gain is larger at higher voltages this operation was possible without cutting the signal distribution.
The measurements were stopped when the waveforms showed instabilities such as deformation in the shape or delays returning to the baseline or when the noise level was so high that the lowest threshold to reject the noise would have cut into the low values of the signal distribution.
Example waveforms of a non irradiated device are shown in figure 4.18. The current pulse is characterized by a duration of the order of about one nanosecond followed by a second smaller pulse due to some impedance mismatch in the readout chain.
The waveform shape is given by the superposition of the induced current from the drift of the primary electrons and holes and from the secondary holes. The current is enhanced when the primary electrons reach the multiplication layer generating more charge carriers. This will last as long as there are electrons drifting to the cathode, expected to be about 450 ps, at the drift velocity saturation of 96µm/ns [66]. This is the intrinsic rise time of the signal. Afterwards the signal consists of the drifting of the multiplied holes until they reach the opposite electrode. The intrinsic pulse is convoluted with the electronics response function whose contributions come from the
sensor capacitance combined with the amplifier impedance and the bandwidth of the oscilloscope.
(a) (b)
Figure 4.18: Waveforms from sample med,unirr,L1 at a bias voltage of 200 V with a gain of 14 (a) and 235 V with a gain of 35 (b).
The amplitude distribution of selected devices and voltages is shown in figure 4.19 and the histograms are fitted with a Landau-Gauss distribution function to extract the amplitude MPV. Figure 4.19a shows the effect of bias voltage on non irradiated devices. At larger voltages the MPV moves to larger values and at the same time the distribution gets broader. Figure 4.19b instead shows the effect temperature and irradiation. At larger fluences, despite the larger bias applied, the pulse amplitude is suppressed while at lower temperature the amplitudes tend to be larger although the diode breakdown occurs at lower voltage. However the amplitude distributions of the med,3e14 device are much broader than before irradiation. The suppression of the amplitude of the med,1e15 sample due to a higher irradiation fluence leads to a lower distribution width as well.
The charge, and hence the gain, follow a behaviour similar to the amplitude MPV.
The charge is calculated integrating the waveform from −1 to 4 ns and the gain is obtained dividing the integral by the signal of an equivalent sensor without gain. It was not possible to directly measure a 45µm thick reference because of the low signal to noise ratio, the charge was calibrated by scaling the signal of a 300µm thick diode.
The gain showed run to run variation up to about 30−40% due to the non stable environment of the test beam area in terms of noise and temperature.
In addition to amplitude (4.20a), charge and gain (4.20b) other important param-eters have been extracted from the analysis of the waveforms. Such as the baseline noiseN (4.20c) taken as the RMS of the waveform baseline, the signal to noise ratio S/N (4.20d) taken as the signal amplitude MPV divided by the noise RMS, rise time τ10−90% (4.20e) taken as the time in which the waveform goes from the 10% to the
4.2. LGAD FOR TIMING 75
Figure 4.19: Amplitude distribution of sample med,unirr,L1 for different bias voltages (a) and for the irradiated devices at the maximum bias voltage measured (b).
90% of its own amplitude and the jitterσjitter (4.20f) that from equation 2.20 can be estimated by σjitter∼τ10−90%/(0.8S/N).
It is interesting to notice how the amplitude MPV and the gain grow exponen-tially approaching the breakdown voltage, confirming the 90Sr measurements shown in figure 4.16, while the noise is stable for non irradiated devices and grows slowly for non irradiated ones so that the S/N ratio results enhanced with larger bias volt-age. On the other hand, the jitter decreases with increasing bias as well as the rise time. The larger signal over noise ratio, slew rate and lower jitter all contribute to an improvement on time resolution as shown in equation 2.18.
The measurement of time resolution of thin LGAD devices was the main goal of the measurement at the test beam facility. It was measured by the spread of the time of arrival difference (∆t) between two devices, which contains the contribution from both devices.
The signal from the SiPMs used as timing reference was processed by a CFD returning a digital signal. The time of arrival was taken at the fixed threshold value of 250 mV, about half of the signal amplitude. Instead, for the LGAD devices the analog waveform was sampled in 25 ps bins and stored. An offline CFD algorithm has been used to correct for time walk fluctuation. The threshold was then defined for each waveform as a constant fraction of the signal amplitude. The time of arrival was taken as the time the signal was crossing the threshold value interpolating linearly from the measurement points just above and below the threshold. Other algorithms were tested, including more points, using polynomial fit or spline interpolation but no significant improvement was noticed. The threshold of the CFD algorithm was scanned in steps of 5% from 10% to 90% of the signal amplitude and the optimal value was taken. This value changed depending on the device and on the voltage
Voltage [V]
100 200 300 400 500 600
Amplitude MPV [mV]
100 200 300 400 500 600
Gain
100 200 300 400 500 600
Noise [mV]
100 200 300 400 500 600
Signal-to-Noise Ratio
100 200 300 400 500 600
Rise Time 10-90% [ps]
100 200 300 400 500 600
Jitter Time Resolution 10-90% [ps]
0
Figure 4.20: Overview of waveform parameters for all the measured devices as a function of voltage.
applied. Probably due to the waveform shape variation that makes the rising edge steeper at different points. For the non irradiated devices the optimal value of the CFD threshold was found to be around 80% at low voltages decreasing to 20% at the maximum measured voltage, while for the irradiated devices it stayed between 80 and 90%.
Figure 4.21 shows the ∆t distributions for different devices. The spread of each distributionσtotalis obtained by the standard deviation of a Gaussian fit that contains the contribution of the two devices. For some runs two LGAD devices have been
4.2. LGAD FOR TIMING 77 measured together with both SiPMs to extract the time resolution of each SiPM.
An analysis of all the possible ∆t combinations returned the individual SiPM time resolution asσSiP M1 = 13±1 ps andσSiP M2 = 7±1 ps, the uncertainties contain the statistical contribution of 0.2 ps and a run by run fluctuation of about 1 ps. The time resolution of the LGAD devices is typically evaluated in combination with SiPM2 that showed a better time resolution. The contribution of the SiPM is subtracted fromσtotal assumingσtotal2 =σLGAD2 +σSiP M2 . During the beam tests some runs have been taken with the two sensors irradiated to 3·1014neq/cm2 in a climate chamber at the temperature of−20◦C, for these runs it was not possible to include any SiPM in the readout so that the average time resolution of the two devices is measured as σhLGADi =σtotal/√
Figure 4.21: Time of arrival difference (∆t) distribution between an LGAD and a SiPM. Device med,unirr,L1 at different voltages (a) and irradiated devices at the maximum measured voltage (b) are shown. The LGAD time resolution σLGAD is obtained by the width of a Gaussian fit subtracting the SiPM contribution.
The dependence of the time resolution on the bias voltage and on the gain is
The dependence of the time resolution on the bias voltage and on the gain is